基于改进YOLOv5算法的钢铁表面缺陷检测

根据以往钢铁表面缺陷检测技术的检测效能较低、准确性低的情况，提出一种改进YOLOv5s的钢材表面缺陷检测算法。主要改进为：加入坐标注意力机制(Coordinate Attention，CA)的空洞空间卷积池化金字塔 (Atrous Spatial Pyramid Pooling，ASPP)，扩大模型感受野和多尺度感知能力的同时能更好的获取特征位置信息；加入改进的选择性内核注意力机制(Selective Kernel Attention，SK)，使模型能更好的利用特征图中的频率信息，提升模型的表达能力；将损失函数替换为SIoU，提升模型性能的同时加快模型的收敛。实验数据表明，改进的YOLOv5s网络模型在NEU-DET数据集上的mAP值为78.13%，相比原网络模型提高了2.85%。改进的模型具有良好的检测型性能的同时检测速度为103.9 FPS，能够满足实际应用场景中钢材表面缺陷实时检测的需求。

Steel surface defect detection based on improved YOLOv5 algorithm

Based on the low detection efficiency and low accuracy of previous steel surface defect detection techniques,an improved YOLOv5s algorithm for steel surface defect detection is proposed. The main improvements are: Atrous Spatial Pyramid Pooling (ASPP) with Coordinate Attention (CA), which expanded the model''s perceptual field and multi-scale perception capability while better acquiring feature location information; An improved Selective Kernel Attention mechanism (SK) was added to make the model can better utilize the frequency information in the feature map and improve the model''s expressiveness; replaced the loss function with SIoU to improve the model performance and accelerate the convergence of the model. The experimental data shows that the mAP value of the improved YOLOv5s network model on the NEU-DET dataset is 78.13%, which is 2.85% higher than the original network model. The improved model has good detection performance and a detection speed of 103.9 FPS, which can meet the needs of real-time detection of steel surface defects in practical application scenarios.